Inflammatory Signaling in Pancreatic Cancer Transfers Between a Single-cell RNA Sequencing Atlas and Co-Culture

Abstract

Pancreatic ductal adenocarcinoma (PDAC) is a devastating malignancy driven by a heterogeneous tumor microenvironment enriched with cancer associated fibroblasts (CAFs) that influence its overall immunosuppressive composition. We examined inflammation in PDAC by leveraging our existing patient-derived organoid (PDO) model and a novel PDO-CAF co-culture. We first identified induction of major histocompatibility complex class II expression following treatment with interferon gamma. In parallel, we collated an atlas of 6 published single-cell RNA-sequencing datasets (174,394 cells) combining 61 PDAC (142,807 cells) and 16 non-malignant samples. By combining in silico modeling and in vitro PDO co-culture, we define a gene expression pattern of inflammatory processes in epithelial tumor cells. Following computational inferences, we examined interactions between epithelial tumor cells and CAFs, focusing on VEGF-A and ITGB1 pathways. This work, integrating computational and biological approaches, highlights the value of convergence to accelerate our understanding of key drivers of PDAC.

PanIN and CAF Transitions in Pancreatic Carcinogenesis Revealed with Spatial Data Integration

“…We observed that the reduction in this inflammation pattern is gradual with progression from normal ductal cells through LG to HG PanIN lesions. It is possible that downregulation of this pattern may be due to development of immune suppressive CAFs that are already present in PanINs and are functionally similar to those present in PDACs, as demonstrated by Kinny-Koster et al using a co-culture patient derived organoid model(18). In addition, CAFs can create an immunosuppressive TME by producing and releasing cytokines that inhibits immune cell infiltration and factors that provide cancer cells additional growth and survival advantages(21,23).…”

“…Specifically, a second transfer learning method was required to uncover PDAC scRNA-seq patterns from the PanIN ST data. Adding to the learned oncogenic and a normal pancreatic function patterns (Patterns 5 and 2, respectively) in the PDAC atlas (18), our approach identified a third pattern enriched for inflammation markers (Pattern 7) that is downregulated in primary PDAC cells. This reduction in inflammatory signals may be one mechanism by which PDACs evade immune recognition, as tumor cells will be producing less immune chemoattractants.…”

“…We obtained scRNA-seq data for pancreatic epithelial cells from an atlas of 29 tumor samples and 14 non-cancerous samples collated from Peng et al and Steele et al as described in Kinny-Koster et al(18). We inferred cellular phenotypes in the epithelial cells using CoGAPS (R, version 3.5.8)(48,49) to infer 8 patterns on the log transformed expression values.…”

“…The PDAC Atlas from six previously published scRNA-seq datasets includes gene expression profiles from 25,442 epithelial ductal cells (Cancer: 14,589 cells; Normal: 7,561 cells; Normal_Tumor_Adjacent: 2,375 cells, Unspecified: 917 cells) (Supplemental Figure 8A). These epithelial populations were analyzed using CoGAPS (48,49) that learnt eight transcriptional patterns associated with biological processes (18). To study the representation of the patterns learned in the PDAC atlas across PanIN stages, the patterns were projected onto epithelial spots from the ST data (N = 623 spots; normal = 254, LG = 110, HG = 159) using projectR (17,19).…”

“…Here, we report the successful integration of CODA with FFPE ST data through transfer learning of the cellular annotations from CODA to the stained images in Visium to enhance pancreas cell type annotation by spatial spots and robustly examine cell-type specific molecular phenotypes in PanIN lesions. To further expand our findings and further delineate the transitional pathways that drive normal cell to PanIN to PDAC development, we applied a second transfer learning approach, projectR (17), to integrate CODA and ST data with scRNA-seq data from prior reported studies that have been entered into a novel PDAC atlas (18). Applying transfer learning to multiple types of data provides the opportunity to create a model of PDAC initiation and progression to advanced disease.…”